Phylogenetic analysis of metastatic progression in breast cancer using somatic mutations and copy number aberrations

Abstract

Several studies using genome-wide molecular techniques have reported various degrees of genetic heterogeneity between primary tumours and their distant metastases. However, it has been difficult to discern patterns of dissemination owing to the limited number of patients and available metastases. Here, we use phylogenetic techniques on data generated using whole-exome sequencing and copy number profiling of primary and multiple-matched metastatic tumours from ten autopsied patients to infer the evolutionary history of breast cancer progression. We observed two modes of disease progression. In some patients, all distant metastases cluster on a branch separate from their primary lesion. Clonal frequency analyses of somatic mutations show that the metastases have a monoclonal origin and descend from a common 'metastatic precursor'. Alternatively, multiple metastatic lesions are seeded from different clones present within the primary tumour. We further show that a metastasis can be horizontally cross-seeded. These findings provide insights into breast cancer dissemination.

Figure 1. Phylogenetic inference from SNVs and CNAs.

Fully clonal somatic CNAs of chromosome 3 for (a) the primary tumour, (b) the liver and (c) the ovarian metastases of patient 2/57. The tracks are in descending order, the B Allele Frequency (BAF), integer copy number (CN) state and log₂ ratios. The phylogenetic reconstruction is displayed in d, where arrows indicate aberrations with annotations besides detailing coordinates and event type. Convergent evolution is exemplified by the focal amplification of region d in a,b but a broad gain of c–e in c, in all three cases leading to copy-neutral loss of heterozygosity of region d. The CNAs are colour coded with ‘early’ events in blue, ‘late’ events in orange, and diploid regions not contributing to the phylogenetic tree in black. (e) Concordant phylogeny obtained from tier-3 SNVs and (f) ancestral state reconstruction for the same samples. The scale bars in (d,e) represent one CNA and 10 SNVs, respectively.

Figure 2. Reversions in SNVs are explained by underlying CNAs.

(a,f) ‘Early’ and ‘late’ tier-3 SNVs, respectively, which were predicted to be reversions in the metastasis to the pylorus of patient 8/82. These were clustered on chromosome 1p and 17p. (b–d, g–i) Fully clonal somatic CNAs of chromosome 1 and 17, respectively, for the primary tumour, the pylorus and the liver metastases ordered according to their genomic coordinates. In each panel, the tracks displayed are in descending order, the BAF, the integer based estimation of CN and the log₂ ratios. (e) Heat map representing the ancestral state reconstruction. The loss of heterozygosity at chromosome arm 1p and 17p in M2 explains the absence of these mutations.

Figure 3. Phylogenetic reconstruction of breast cancer progression in patient 7/67.

(a) Ancestral state reconstruction of tier-3 SNVs with the anatomic location of the profiled lesions is depicted in b. (c) Combined phylogenetic tree obtained from CNAs and SNVs and (d) pairwise comparisons of clonal frequencies of tier-4 SNVs. The branches of the phylogenetic tree are labelled 1–5 and the location of these mutations in pairwise comparisons is indicated in d. (e) Schematic representation of the pairwise comparison of two fictitious samples. Mutations in i, ii and iii are fully clonal being either common to the two samples and thus inherited from their parental lineage or private to either one. Mutations in iv and v are private and subclonal to either samples. They are expected to have occurred after the divergence of the two lineages and after mutations located in ii and iii, respectively. Mutations in vi and vii are shared between the two samples but are fully clonal in one and subclonal in the other. If the two samples share a common parental origin, these mutations are incompatible with fully clonal mutations occurring in ii and iii, respectively. A possible scenario explaining their occurrence is that vi and vii are mutually exclusive and that sample #1 seeded #2 giving rise to vi or vice versa for vii. The subclonal frequencies could then be explained by intra-tumour heterogeneity in the tumour mass. Alternatively, mutations in vi and vii could find their origin in horizontal reseeding from a third sample.

Figure 4. Phylogenetic reconstruction of breast cancer progression in patient 2/57.

(a) Anatomical representation of tumour lesions profiled, (b) combined phylogenetic tree obtained from CNAs and SNVs, and (c) pairwise comparisons of clonal frequencies from tier-4 SNVs. The branches of the phylogenetic tree are labelled 1–6 in b and the location of these mutations in pairwise comparisons is indicated by the corresponding label in c. Mutations in segment 2 at full clonal frequencies in M3 and subclonal frequencies in M2 indicate horizontal seeding, highlighted by the red segment 2 in b. A heuristic interpretation of the different possible scenarios is given in d. Only three mutations in total were in the unexplained configuration *, one in 5, one in 6 and 12 in configuration 2. Excluding mutations in the configuration *does not influence the topology of the phylogeny. The numbers in parentheses in d give the percentage of all tier-4 SNVs.

Figure 5. Combined phylogenies representing metastatic progression across eight patients.

(a) Phylogenies of early stage patients who underwent primary surgery followed by systemic treatment and (b) phylogenetic trees obtained from advanced stage treatment naïve and de novo metastatic patients. The same colour code as in previous figures is used to depict ‘early’ and ‘late’ events. For visual purposes, all the trees were globally rescaled such that the trunks of the trees have the same length. The scale bars at the bottom represent 10 SNVs and provide an indication of the original length of the trees. For patients 4/71 and 9/68, the primary tumour samples removed at surgery were exome sequenced and putative tier-1 somatic mutations were further validated by Sequenom MassARRAY and ultra-deep amplicon sequencing. However, the corresponding SNP arrays showed that these samples had CCFs below the set threshold of 30% for phylogenetic reconstruction. Nonetheless, for these two particular samples, tier-3 SNVs were included in the construction of the phylogenetic trees for SNVs on account that the lesions had been removed several years prior to the diagnosis of distant relapses and autopsy. Similarly, for patient 10/80, the primary tissue samples did not pass the filtering criteria of tier-3 level. The thickness of the branches leading to these nodes is therefore irrelevant. These are displayed in grey.

Figure 6. Dynamics of genomic alterations during metastatic progression.

Normalized phylogenetic distance for each sample profiled. These are obtained as the ratio of the path from the common ancestor node to the given sample relative to the trunk of the tree for (a) SNVs and (b) CNAs. (c,d) Correlation of the average phylogenetic distances with overall survival for SNVs and CNAs, respectively.

Figure 7. Distribution of substitutions during metastatic progression.

Frequency of the different types of substitutions for tier-3 SNVs in patients (a) 8/82 and (b) 10/80. These are grouped as ‘early’ and ‘late’ according to their occurrence in the respective phylogenetic trees.